(Master’s Thesis in Finance)
M. Prieˇckov´aa,∗
aUniversity of Groningen, Faculty of Economics and Business
Abstract
In this thesis, I study the relationship between environmental performance (EP) and financial performance. I create 5 mutually exclusive portfolios based on the level and the change of environmental score. I regress value-weighted average monthly portfolio returns using capital asset pricing model (CAPM) and Carhart 4-factor model. I also study downside risk and risk-return trade-off of the portfolios using semideviation, and Sortino ratio and Sharpe ratio. The findings indicate that the return of the portfolio of EP leaders does not significantly differ from the return of the portfolio of EP laggards. I addition, the portfolio of EP leaders significantly underperforms the market. Except for the case the portfolios are created based on the change in EP score and at the same time Carhart 4-factor model is used when the portfolio of EP leaders does not significantly differ from the market. Even though the downside risk of the portfolio of EP laggards is higher, the risk-return trade-off ratios are better compared to the portfolio of EP leaders in case the portfolios are created based on environmental score. For the portfolios created based on environmental score change, the portfolio of EP leaders has slightly lower downside risk and slightly better risk-return trade-off ratios than the portfolio of laggards.
Keywords: Environmental performance, Stock returns, Portfolio study, European market, Semideviation, Sortino ratio, Sharpe ratio
JEL classification:G11, G12, Q56
II would like to thank to my supervisor, Prof.dr. L.J.R. Scholtens, for help with topic choice, useful feedback and helpful
suggestions. I would like to thank to my friend, Matej Baˇst´ak, for technical support.
∗Student number: 2363836
Email address: maria.prieckova@gmail.com (M. Prieˇckov´a)
1. Introduction
In recent years so-called socially responsible investing (SRI) is a “hot” topic on financial markets. In my thesis, I focus on the environmental dimension of the broad topic of corporate social responsibility as the demand for “green” products and “green” investment is increasing because of more and more debated issues of climate change and environmental load, and becomes more and more important driving factor of the investment decisions. Numerous theoretical and empirical studies in this domain have not found the answer on the question whether it does pay to be “green”, yet. However, this question is crucial for the investors seeking “green” investments because they care about environment protection in order to know whether they should expect a reward or a penalty. And it is crucial also for the investors neutral about environment protection but seeking for higher than market return in order to know whether environmentally screened investment brings them such a return. Thus, I will study the relationship between environmental and financial performance and examine whether the portfolio of companies with high environmental per-formance has higher stock return compared to the market, and compared to the portfolio of companies with low environmental performance, too.
The contribution of my thesis relies in two ways of the environmental performance definition. Besides widely studied relationship between the level of environmental performance and stock returns I also study the change in environmental performance and stock returns. This point of view captures the link between evolution in environmental performance and evolution in stock price. Another contribution is the inves-tigation of downside risk and risk-return trade-off which is neglected in most of studies. My thesis also contributes to the existing literature by focusing on less studied European market with an extended period of time lasting from February 2003 to March 2013.
In order to study my hypothesis I will use so-called portfolio method. First, I will 5 create mutually exclusive portfolios of companies based on their environmental performance. Second, I will regress value-weighted monthly average returns of each portfolio using 1-factor capital asset pricing model (CAPM) and 4-factor model ofCarhart(1997) incorporating size factor, value factor and momentum factor into CAPM. Finally, I will analyze the downside risk of portfolios using semideviation, and the risk-return trade-off of portfolios using Sortino ratio and Sharpe ratio. I will repeat this methodology for both environmental performance definitions, first for the level of environmental performance (measured as ESG ASSET4 envi-ronmental score), and second for change in envienvi-ronmental performance (measured as ESG ASSET4 yearly environmental score change).
In next section I discuss empirical research up to date. In Section3, I develop and formulate research hypotheses. Then I describe data in Section4, and I present the research methodology in Section5. Before concluding in Section7, I analyze the results in Section6.
2. Literature review
In the last three decades many academics and researchers studied the relationship between environ-mental and financial performance of companies. The main question they focus on is whether higher envi-ronmental performance of a firm enhance firm‘s future financial performance by generating some benefits such as lower cost of capital (Bassen et al.,2007;Figge and Hahn,2006), higher returns (Rao,1996; Ya-mashita et al.,1999;Ziegler et al.,2007) or a competitive advantage (L´opez-Gamero et al.,2009;Wagner and Schaltegger,2004;Porter and van der Linde,1995).
Although interest in this domain is very large, results of studies are not consistent. Some authors find positive relationship between environmental and financial performance (King and Lenox, 2001; Derwall et al.,2005) while others find that “it does not pay to be green” (Filbeck and Gorman, 2004) or that en-vironmental performance does not influence firm‘s financial performance (McWilliams and Siegel,2001; Mahapatra,1984). Apart from those results,Wagner and Schaltegger (2004) andWagner (2005) find so-called inverse U-shaped relationship which means that environmental performance has a positive impact on financial performance until a given level from which costs of environmental responsibility outweigh the benefits and relationship becomes negative.
The empirical evidence inconsistency can be caused by differences among studies such as different time periods and countries, different sources of data or different methodologies. Besides those usual reasons we observe the field specific issue - measurement of environmental performance. First of all, it is worth to say that measuring environmental performance is very difficult and still not a unified process. Thus using different proxies for the environmental performance measure leads to the divergent results and also causes difficult comparability between studies (Wagner,2001).
At the beginning, authors use firm’s information about the pollution control (Spicer,1978) and the pollu-tion control expenditures (Mahapatra,1984) as the environmental performance proxy.Darnall et al.(2008) andL´opez-Gamero et al.(2009) use data gathered from companies’ managers to measure environmental performance. The limitation of such a proxy is that managers are used to exaggerate their results when self-evaluating and that they use the different benchmark to evaluate firm’s environmental performance. Other researchers calculated their own environmental performance proxy. Wagner (2005), when studying European pulp and paper industry use “SO2emissions, NOx, emissions, COD emissions, total energy input,
and total water input, all per ton of paper produced” to obtain firm’s environmental performance. In a similar wayHart and Ahuja(1996) construct the environmental proxy from the emission reduction data or Earnhart and Lizal(2006) construct the environmental proxy based on the amount of air emission. Environ-mental ratings and indices are also widespread source of environEnviron-mental performance measure used mostly in portfolio studies (Blank and Daniel,2002;Filbeck and Gorman,2004;Derwall et al.,2005).
that companies with better pollution-control have also higher profitability. Second,Mahapatra(1984) stud-ies companstud-ies listed on New York Stock Exchange. Contrary to (Spicer, 1978), his results suggest that pollution-control expenditures level does not influence average market return on common stocks.
During past 30 years, 3 main types of studies investigating relationship between firm‘s environmental and financial performance have developed. First, event studies focus on short-term response of financial performance to an environmental event, second, multiple regression studies investigate long term relation-ship on a firm level, and third portfolio studies examine long term relationrelation-ship on the level of mutually exclusive portfolios (Wagner,2001).
2.1. Event studies
Event studies are based on the Efficient Market Theory (Fama, 1970) which says that the price of a listed company includes both current and expected financial performance in the market valuation. If the stock returns after a positive or negative environmental event significantly differ from the expected returns, market attributes this change in firm value to a given positive or negative environmental event (Wagner, 2001). A limitation Wagner (2001) points out is that event studies “can only be based on stock market-based company performance data”, which means that they use only difference between expected and actual return to study influence of an environmental event on firm‘s stock return. In TableAppendix AI show brief review of event studies.
Pioneering event study (Shane and Spicer,1983) investigate the market reaction to the disclosures of control records between 1970 and 1977. They examine 2 disclosures for each of following industries: paper, power, steel, and oil. They find that, on average, firms experience relatively large negative abnormal returns on the two days immediately prior to the disclosure. The results also suggest that companies with lower pollution control ranking have significantly larger negative abnormal return than companies with higher ranking. Those findings suggest positive relationship between environmental and financial performance.
Similar results are found byRao(1996). He studies 1-month abnormal return of 14 U.S. multinational companies after the Wall Street Journal release of information about pollution incidents during years 1989 – 1993. He finds that firms have significantly negative monthly abnormal return from 12 months before the announcement to 6 months after the announcement.Hamilton(1995) examines the impact of the Toxic Release Inventory publication on 1-day abnormal returns. Concerned companies seem to have significantly (with the confidence interval of 1%) negative abnormal returns on the announcement day and during 5 days from the announcement day.Hamilton(1995) also pointed out that the higher number of different chemical submissions firm reports the larger is the drop in its stock value.
Klassen and McLaughlin (1996) study 96 firms encountering a positive environmental performance event (winning an award) and 16 firms encountering a negative environmental performance event (oil or chemical crisis, explosion etc.). They find evidence that 3-day cumulative abnormal return is significantly (with confidence level of 5%) higher for companies which won an environmental award and significantly (with confidence level of 1%) negative for companies which faced a recent environmental crisis.
with poorer environmental conscientiousness score. Still, for firms with high environmental conscientious-ness score no significantly positive returns are found even when considering long-term relationship.
Besides the investigation of the effect of an environmental event on the short-term abnormal stock return, event studies can focus on the accounting performance measures. Lo et al.(2012) study 3 financial ratios, namely return on assets (ROA), return on sales (ROS), and sale on asset (SOA), of firms from U.S fashion and textile industry. They investigate reaction of those ratios to environmental management systems (EMS) adoption. As firm needs to meet several criteria before formal adoption of EMS which is motivational for them, the authors study period from 2 years before adoption of EMS to 1 year after adoption of EMS. The results of their study suggest that adoption of an EMS has the positive and significant impact on ROA and ROS. Coefficients for both ratios are significant with confidence interval of 1% for period from 2 years before adoption to the year of adoption of EMS. On the other side, the impact on SOA during this period remains insignificant. Only a small positive and significant with confidence interval of 10% change of SOA appears in the year after adoption of EMS.
2.2. Multiple regression studies
Multiple regression studies focus on the long-term relationship between company‘s environmental and financial performance. Authors usually include one or more environmental performance proxies into the regression equation while controlling for other firm specific characteristics. As in the case of event studies, authors investigate either market performance (Thomas, 2001;Ziegler et al., 2007; Darnall et al., 2008) or accounting performance (Hart and Ahuja, 1996; King and Lenox, 2001; Wagner, 2005; Horv´athov´a, 2012). However this method calls for very strong model of linkage between variables and a large number of observations (Wagner,2001). I present summary of multiple regression studies in TableAppendix B.
Thomas (2001) studies relationship of 3 environmental proxies and the excess stock returns on the United Kingdom (UK) market during the 1985 – 1997 period. Although she finds that the environmen-tal policy adoption slightly enhance the financial performance, two other measures – prosecution by an environmental agency, and environmental training of staff – have no effect on firm‘s excess returns. The results ofDarnall et al.(2008) also suggest positive the relationship between environmental and financial performance. They test whether increased level of environmental management system (EMS) comprehen-siveness influence firm‘s profitability and growth. Their findings indicate that more comprehensive the EMS is adopted higher is the business performance of the firm.
Hart and Ahuja(1996) study how the return on equity (ROE), ROA, and ROS react on the firm‘s emis-sion reduction. Their findings suggest positive and significant (with confidence interval of 10%) relationship between all 3 financial ratios and firm‘s improved environmental behavior. Specifically ROA and ROS are benefited 1 year after the reduction of emission, and it takes 2 years to enhance ROE.Horv´athov´a(2012) obtains similar results from Czech market during the 2004 – 2008 period. In her study ROE an ROA seem to be positively related to emission reduction variable lagged 2 years.King and Lenox(2001) do not study change in emissions but its level using panel data analysis. They regress Tobin’s q1on total level of emis-sions produced by firm and by industry, and on firm‘s emisemis-sions relative to average emisemis-sions in the firm‘s industry. Contrary toZiegler et al.(2007) they find that industry emissions do not influence financial perfor-mance of a firm. However, both measures of the firm emissions have slightly significant (with confidence interval of 10%) positive effect on financial performance.
Wagner (2005) studies European pulp and paper industry using panel data analysis and hypothesizes the inverse U-shaped relationship between environmental and financial performance. He uses firm‘s infor-mation about various emissions, and energy and water input (all per ton of paper produced) to calculate an environmental index. Then he regresses return on capital employed (ROCE), ROS, and ROE on the calculated index, and on the square of the calculated index controlling for the asset turnover ratio, the debt-to-equity ratio, firm size, the square of firm size, and the sub-sector and country dummies. Although the coefficient for index is positive and the coefficient for square of index is negative suggesting inverse U-shaped pattern of relationship both coefficients are statistically insignificant. Only for fixed effect model the coefficient of environmental index is significant while the one of square of environmental index remains insignificant.
2.3. Portfolio studies
Portfolio studies, as multiple regression studies, examine the long-term relationship between environ-mental and financial performance. The essential difference is not to include an environmental proxy into the regression but to create 2 or more mutually exclusive portfolios based on environmental performance and study whether financial performance (accounting or market measured) differs across portfolios. If yes, financial performance is influenced by environmental performance. The drawback of the method is that only an average performance across portfolio is assessed (Wagner,2001). Review of portfolios studies is in TableAppendix C.
Cohen et al.(1997) rank S&P500 companies based on rating created by 8 different environmental perfor-mance measures and compare the balanced portfolio of high ranked companies with the industry-balanced portfolio of low ranked companies. Their results suggest that environmental leaders perform as well as (or better) environmental laggards in both accounting (ROA, ROE) and market (stock returns) fi-nancial performance measures. Thus, an investor who cares about environment protection can create an industry-balanced portfolio of environmental leaders and expect the return at least as the return of S&P500 index. AlsoGuerard(1997) “concludes that portfolios derived from a socially screened investment universe did not perform differently from those obtained from an unscreened set”, as stated inDerwall et al.(2005).
Filbeck and Gorman(2004),Blank and Daniel (2002), andDerwall et al.(2005) also use an environ-mental score to separate companies into environenviron-mental leaders and laggards. In all 3 cases authors do not
calculate the scores but use a score published by rating agencies specialized in ESG research.
On the one handFilbeck and Gorman(2004) find that the portfolio of “more compliant” firms has lower return than the portfolio of “less compliant” firms and both of them underperform the market. Although this pattern is observable the difference between two portfolios are insignificant. Only return of portfolio of “more compliant” firm is significantly (with confidence interval of 5%) lower than the market return which indicates that in U.S. electric utility industry it does not pay to be “green”. Regarding study of Sharpe ratio, the findings suggest that the portfolio of “more compliant” firm brings higher return per unit of risk (measure as historical standard deviation) than the portfolio of “less compliant” firms. Still both portfolios have lower Sharpe ratio than the market.
On the other hand,Blank and Daniel(2002) andDerwall et al.(2005) find significant and positive rela-tionship between higher environmental score and higher return. First,Blank and Daniel(2002) analyze the companies with available Innovest rating score (environmental score). They study the returns of portfolio of top ranked companies and the return of portfolio of all ranked companies, S&P500 index return, and the return of portfolio of bottom ranked companies. They find that the portfolio of top ranked companies has higher annualized return for 4-years period (1997 – 2000) than all three other portfolios. The difference is the highest in case of the portfolio of bottom ranked companies. The authors also analyze risk-return trade-off of S&P500 index compared to the portfolio of top ranked companies. The portfolio of top ranked companies has higher Sharpe ratio so has higher return on a given unit of risk. Thus, also the analysis of Sharpe ratio indicates that the investment in “green” companies does pay.
Second,Derwall et al.(2005) study U.S. firms based on Innovest eco-efficiency score (similarly toBlank and Daniel,2002) but they use more sophisticated study method – multivariate regressions of CAPM and Carhart 4-factor model. They test for difference in Jensen‘s alphas of 2 portfolios, 30% high-ranked and 30% low-ranked companies. The results of both regressions (CAPM and 4-factor model) indicate that the portfolio of high-ranked companies has higher return than the portfolio of low-ranked companies. However, while in case of CAPM the difference is not significant, in case of 4-factor model the difference is slightly significant (with confidence interval of 10%).
Ziegler et al.(2011) use the same models for their study asDerwall et al.(2005). They screen companies based on 2 different characteristics; one is the disclosure of a climate impact statement and the other is the release of carbon reduction information. They regress monthly excess stock return using CAPM and Carhart 4-factor model for the portfolio with and the portfolio without a given characteristic, and for the portfolio with both and the portfolio with none of characteristics. The findings indicate that for European market in years 2004 –2006 and U.S energy market the portfolio of companies with both of characteristics has higher return that the portfolio of companies with none of characteristics. However, for European market during 2001 – 2006 the portfolio of companies with climate impact statement (or with both characteristics) has lower and slightly significant (with confidence interval of 10%) return than the portfolio without climate impact statement (with none of the characteristics).
2.4. Contribution of my study
Based on the reviewed studies I consider that there is a lack of studies focusing on European market. Only 2 studies use data from more European countries (Ziegler et al.,2007, 2011), 3 studies focus on 1 or few European countries (UK market (Thomas,2001), Czech market (Horv´athov´a,2012), and German, Italian, Dutch and UK market (Wagner, 2005)). This number is certainly low compared to 14 studies exclusively about U.S. market. Therefore, I investigate the relationship between environmental and financial performance in Europe. My study fills the gap in portfolio studies, where the lack of studies of European market is even more evident. I also use in my study an extended study period compared to other studies.
The contribution of my study is twofold. First, I investigate not only the return of environmentally screened portfolios but I also analyze the downside risk and the risk-return trade-off of the portfolios which more accurately capture investor‘s perspective. Second, besides usual creation of the portfolios based on the level of environmental performance I also create the portfolios based on the change in environmental performance. This perspective links evolution of environmental performance with evolution of financial performance.
3. Hypotheses development
For last thirty years, the relationship between environmental and financial performance is widely de-bated in the academic circles. However, the answers on the essential questions remain still unrevealed.
On the one hand, the causality between environmental and financial performance is still questionable. Does high environmental performance enhance financial performance or does higher financial performance allow more spending on environment protection followed by improved environmental performance? Al-though most of authors examine the former question and their results indicate this direction of the causality, Earnhart and Lizal(2006) focus on the latter when studying Czech firm during 1993 – 1998. Their find-ings suggest that better financial performance endorse higher environmental performance.Scholtens(2008) studies causality between financial performance and social performance on a sample of U.S. companies dur-ing years 1991 – 2004. Although the results, in general, indicate that financial performance precedes social performance, the environmental dimension of social performance tends to have different pattern. Precisely when using Ordinary Least Square (OLS) analysis with distributed lags it appears that the impact between the financial returns and environmental performance is not very different whether the financial returns are the dependent or the independent variable. Using Granger causality test for 3 years the results suggest that causality runs from the environmental concerns to the financial returns while the results of Granger causal-ity test for 5 years indicate causalcausal-ity running in both directions. Thus, causalcausal-ity between environmental and financial performance is still ambiguous.
end-of-pipe approach2 towards environmental issues. Consequently, the costs linked to the removal or to the elimination of an environmental crisis are considered as the extra costs.
Second, the “revisionists” proclaim that improved environmental performance can be a competitive ad-vantage. The end-of-pipe is very costly approach to environmental performance improvement and overlook the inefficiencies in the production process which can be eliminated. For example incomplete use of re-sources and poor processes lead to unnecessary waste and the increased costs at the level of company, or defected and low quality products leads to additional use of energy and other resources at the customer level (Cohen et al.,1997). Porter and van der Linde(1995) suggest focusing on pollution prevention. This approach launches the decrease in material and resources usage and increase the production and distribu-tion process efficiency. Jointly with product innovation leading to not only “green” production but also to “green” usage and disposal of product, all that can lower resources requirement and encourage firm‘s reputation as well as investors‘ and customers‘ trust.
Third, the view synthesizing two previous theories says that the relationship between environmental and financial performance has an inverse U-shaped form. This means that “revisionists‘” arguments are valid until an optimal level from which the costs start to outweigh benefits and an added environmental improve-ment leads to deterioration of firm‘s profitability. Wagner(2005) incorporate in the regression model not only environmental index of a firm but also the square of environmental index. He finds that financial per-formance is influenced positively by environmental index and negatively by square of environmental index. These findings support the inverse U-shaped pattern of environmental and financial performance.
In the research available till nowadays the majority studies treats environmental performance as in-dependent variable and financial performance as in-dependent variable (Margolis and Walsh (2001) stated inScholtens (2008)). The results of majority of these studies suggest that higher environmental perfor-mance enhance financial perforperfor-mance of a firm.Horv´athov´a(2012) investigates results of 37 studies about relationship of environmental and financial performance and concludes that 55% of studies find positive re-lationship, 30% show no or insignificant relationship and only 15% find negative relationship. In addition, I suppose that positive relationship exists not only because of strong argument of ecological production process saving resources and consequently lowering costs but also because of increasing investors‘ demand for “green” investments and increasing consumers‘ taste for “green” products3.
When studying influence of environmental performance on firm‘s financial performance authors use accounting (e.g. ROA, ROE, ROS) or market (stock returns) measure of financial performance. The ac-counting measures represent past and already realized economic performance while market measures reflect expected future firm‘s performance. This means that stock returns equal to the all discounted future cash flows to the shareholders. Besides this difference, market measures predominate as they are not subject to country specific General Accepted Accounting principles (GAAP) and so-called creative accounting4, and are more comparable across companies. Based on the arguments above I consider the total returns, encom-passing stock prices and also cash flow to the investors, as an accurate measure of financial performance.
Regarding environmental performance measure, 3 possible sorts of proxies are in question. First, total or relative amount of the specific pollutants. For exampleHorv´athov´a(2012) uses excess amount of S O2, NOx
2approach focused on the solutions after an environmental issue arise and not on the prevention
3Eurosif, European SRI study, http://www.eurosif.org/research/eurosif-sri-study/sri-study-2012, accessed June 19, 2013 4Accounting procedures (mainly realization and classification) which leads to the manipulation of accounting results and are
and Chemical Oxygen Demand (COD) over a threshold value. Second, some authors calculate their own environmental performance indices based on few chosen features.Wagner(2005) use 3 pollutant emissions (S O2, NOx, COD) and 2 sorts of resources (energy and water input) to calculate the firm environmental
score. Similarly, Cohen et al. (1997) calculate the environmental score based on 8 firms characteristics (number of environmental litigation proceedings, number of oil spills, volume of oil spills, ect.). Third possibility is to use an environmental performance score or index (Filbeck and Gorman,2004;Blank and Daniel,2002;Derwall et al.,2005) available in Environmental, Social, and Corporate Governance (ESG) focused databases (Innovest, The Investor Responsibility Research Centre, Sarasin & Cie, ESG-ASSET4). As the first and the second type of proxy include only several environmental performance characteristics, in my opinion, those are suitable for studies focus on one or few industries with similar environmental load. On the contrary, I consider an environmental performance score or index to be suitable for the studies in-cluding divers industries as it encompasses more environmental performance characteristics and eliminates the possibility of overlooking an important one.
When studying long-term relationship between environmental and financial performance two methods of studies are in question. Multiple regression analysis and portfolio analysis, both have some pros and cons. While the former catches how influential the dependent variable is in the interaction with all independent variables (e.g. uniformly negative, uniformly positive, parabolic or neutral), the latter does not catches this relationship specification but indicate whether one portfolio differs from another (Wagner,2001). Wagner (2001) also point out that multiple regression analysis requires very sound theoretical model linking vari-ables but still with the risk of interdependence between included independent varivari-ables. Although portfolio analysis compares characteristics of portfolios this can be an advantage as it allows “establishing more clearly systematic differences in economic performance over a larger magnitude of environmental perfor-mance”(Wagner, 2001) and “weakening possible influences of firm-specific variances on the estimation results due to its joint analysis of a group of corporations”(Ziegler et al.,2011). I consider portfolio anal-ysis to be suitable study method based on two advantages mentioned above. Thus, I hypothesize that the portfolio of environmental leaders outperforms the portfolio of environmental laggards. I also hypothesize that portfolio of environmental leaders outperforms the market. In order that these hypotheses are testable I formulate them in the following way:
HA0: The portfolio of companies with high environmental score does not have significantly different
return from the portfolio of companies with low environmental score.
HB0: The portfolio of companies with high environmental score does not have significantly different
stock return from the market return.
As investors regard mainly the change in their wealth (we consider stock returns and not the stock price when determining financial performance of a company) I also consider that they have the same attitude towards environmental performance and consequently they regard the environmental performance evolution in time. I hypothesize that the portfolio of companies whose environmental score increase outperforms both portfolio of companies whose environmental score decline and the market. I formulate these hypotheses as follow so that they are testable:
HC0: The portfolio of companies with increasing environmental score does not have significantly di
ffer-ent return from the portfolio of companies with decreasing environmffer-ental score.
ffer-ent return from the market return.
Markowitz‘s Modern Portfolio Theory (MPT) (Markowitz,1952) says that not only return is important when investing but also risk should be taken into consideration. In the literature oriented on the relationship between environmental and financial performance only few authors study both (Blank and Daniel, 2002; Filbeck and Gorman, 2004). Markowitz‘s MPT state that investment with a given return and the lowest risk or, alternatively, the investment with a given risk and the highest return should be chosen. MPT also says that investment with higher risk is rewarded (for this risk) with higher return. Empirical studies show that when considering risk of stock returns, downside risk is a key factor as it catches only risk of a loss and not the risk of a gain (in comparison with standard deviation in which also risk of gains is included). Downside risk is important for 2 reasons: first, only this risk is relevant from the point of view of an investor, and second, the returns may not follow normal distribution (Markowitz(1959) stated inNawrocki (1999)). Widely used downside risk measure is semideviation which calculation includes only variation below a threshold value, usually one of mean return, risk free rate, market return or an return rate specified by investor. Besides semideviation, lower partial moments (LPM) are used. LPM are used if more than one utility function is considered while semivariance is used if only one – quadratic equation – represents investor‘s utility function (Nawrocki,1999). Thus, I consider semideviation to be appropriate downside risk measure for my study.
As portfolios‘ returns and risks differ and consequently they are not easily comparable reward-to-variability ratios are used as the measures of trade-off between risk and return. These ratios basically represent return over a threshold value per unit of risk allowing easy comparison across portfolios. Higher the ratio is better the trade-off between risk and return is. So that I can compare risk-return of portfolios I investigate also two reward-to-variability ratios. First, Sharpe ratio (Sharpe, 1966) in which standard deviation is used, and second, Sortino ratio in which semideviation is used as the risk measure.
4. Data
4.1. Environmental data
In my study, I use Environmental score (ES) from Thomson Reuters ASSET4 database (used also in Ziegler et al. (2011)) as the environmental performance measure. Thomson Reuters ASSET4 database is a world-wide leader in providing impartial and measurable extra-financial information of firms. This database provides key scores of more than 3200 global companies’ environmental, social and corporate governance (ESG) performance. All information about environmental score is acquired from website of Thomson Reuters ASSET4 and Datastream.
Source of information used for assessment is strictly limited for publicly available data encompassing sustainability/ CSR reports, company websites, annual reports, proxy filings, non-governmental organiza-tions, news of all major providers and in case of CO2data the source is Carbon Disclosure project. As the
annual reports and the CSR reports of companies are published yearly also environmental score is available on a yearly basis from fiscal year 2002 to fiscal year 2012 and is published in most of the case in Febru-ary. For few companies environmental score is not published in February because of the mismatch between firm‘s fiscal year and calendar year.
Environmental score covering various dimensions of company‘s business and its impact on the envi-ronment and sustainability is a reliable and objective measure of a firm’s envienvi-ronmental performance. In addition, Thomson Reuters ASSET4 use only publicly available information which is consistent with the investor perspective of my study as the investor has also access only to publicly available data (Ziegler et al., 2011).
4.2. Financial data
Besides ESG data, Thomson Reuters database is a provider of financial data, too. I use Thomson Reuters DataStream as the source of company specific financial data, namely Total Return Index, Market Value, and Common/Shareholders’ Equity, all with monthly frequency for period from February 1, 2002 to April 1, 2013. All information and formulas about company financial data are acquired from website of Thomson Reuters ASSET4 and Datastream
Total Return Index (TRI) shows a theoretical growth in value of a shareholding over a specified pe-riod, assuming that dividends (gross) are re-invested (ignoring tax and re-investment charges) to purchase additional units of an equity or unit trust at the closing price applicable on the ex-dividend date. TRI is calculated as follows:
RIt= RIt−1× Pt/Pt−1, (1)
except when t is ex-dividend date when TRI is calculated as:
RIt = RIt−1× (Pt+ Dt)/Pt−1, (2)
where Pt is price on ex-date, Pt−1 is price on previous day, and Dt is dividend payment associated with
ex-date t.
Market Value (market capitalization) is the share price multiplied by the number of ordinary shares in issue. The amount in issue is updated whenever new tranches of stock are issued or after a capital change. Market Value is kept in millions of euros.
Common/Shareholders’ Equity represents common shareholders’ investment in a company. It includes several components among which for example Common stock value,Retained earnings, Capital surplus, Capital stock premium. Common/Shareholders’ Equity is kept in thousands of euros.
Besides financial information specific for each company I use risk free rate and market return. I consider 3-month EURIBOR to be risk free rate. EURIBOR is publicly available as annualized interest rate thus, in order to have risk free rate in line with firm specific financial information I convert downloaded5rates to
monthly interest rate using following formula:
rmonth = 12
q
1+ ryear− 1. (3)
Next, for the purpose of market return calculation I consider STOXX Europe 600 Index to properly reflect the return of the European market. The STOXX Europe 600 Index encompasses 600 large, mid and small capitalization companies across 18 European countries (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and the United Kingdom) and monthly data are available online6.
4.3. Final dataset and descriptive statistics
In my study, I examine relationship between environmental and financial performance on the European market and I include publicly listed companies from all European Union member countries plus companies from Norway and Switzerland which is 29 countries and more than 10,000 stocks. Then I include only companies for whose environmental score is available at least once during the 2002 – 2012 period and I exclude duplicates of companies listed at more than one Stock Exchange. Final sample includes 959 companies from 20 countries (Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxemburg, The Netherlands, Norway, Poland, Portugal, Spain, Sweden, Switzerland, and UK)7 and from 10 industries (Basic Materials, Consumer Goods, Consumer Services, Financials, Health Care, Industrials, Oil & Gas, Technology, Telecommunications, and Utilities).
As mentioned above, environmental score is based on publicly available information and is calculated mainly based on annual reports and CSR reports. Thus, environmental score for a given year is, from the investor perspective, available after the end of the year. In order to keep this investor‘s perspective I lag environmental performance with one year. This means that environmental performance for year t is linked with financial performance for year t+1. In other words, at the beginning of year t investor acquires information about environmental performance for year t-1 and based on this information she invests for the rest of the year t or till the next information about environmental performance is available. Therefore, I match environmental performance for years 2002 – 2012 with financial performance for years 2003 – 2013. After merging environmental and financial data I obtain final dataset containing 122 monthly returns (from February 2003 to March 2013) for each company. I use TRI to calculate monthly logarithmic returns of each company:
ri,t = ln(TRIi,t+1/T RIi,t), (4)
where ri,t is return of company i in month t, TRIi,t+1 is Total return index of company i in month t+1, and
TRIi,tis Total return index of company i in month t.
As a company is included in the study only if environmental score is available the number of companies varies across years. In 2002 only 412 companies are included. Then the number of companies increases each year, reaches maximum in 2010 (896 companies), remains over 890 companies in 2011 and drops to 226 in 2012. Detailed evolution of number of companies with environmental score available and yearly average environmental score is inAppendix D.
6http://www.stoxx.com
7Remaining 9 European Union countries (Bulgaria, Cyprus, Estonia, Latvia, Lithuania, Malta, Romania, Slovakia, Slovenia)
Descriptive statistics of companies‘ environmental score and monthly logarithmic stock return as well as descriptive statistics of risk free rate and market return are in Table1. Environmental score is calculated based on sample of 7476 observation (sum of companies with available environmental score in February during years 2002 – 2012) and is relatively stable over time as values vary from 55 to 75. Extreme Kurtosis of stock return is caused by few extreme cases (can be seen also from maximum and minimum figure). Namely, in 5 observations stock return is higher than 1 (100%) and in 35 observations it is lower than -1 (-100%). As those cases are exceptional and I do not study stock returns on firm level I keep in the sample also those observations. In line with portfolio theory, average stock return is approximately equal to market return (STOXX Europe 600) but market return has substantially lower standard deviation (0.04664 for market rate and 0.11074 for stock return).
Table 1: Descriptive Statistics
In this table I show descriptive statistics of environmental score, stock return, return of STOXX Europe 600 (market return), and EURIBOR (risk free rate) for period February 2003 – March 2013. Stock return is monthly logarithmic return calculated based on Total Return Index for each company each month during 2033 – 2013. Return of STOXX Europe 600 is monthly index return and EURIBOR is 3-month annualized EURIBOR transformed to monthly rate using formula rmonth= 12p1+ ryear− 1.
Mean Standard deviation Kurtosis Skewness Minimum Maximum Environmental score 60.93438 29.97379 1.62429 -0.41716 8.99 97.18 Stock return 0.00590 0.11074 41.54259 -1.63883 -4.38241 1.77260 STOXX Europe 600 0.00600 0.04664 4.33480 -0.80623 -0.14211 0.11876 EURIBOR 0.00178 0.00113 2.32890 0.53617 0.00016 0.00431
5. Methodology
5.1. Portfolio creation and return calculation
In the literature, various numbers of portfolios are used to compare performance of environmental lead-ers and laggards. Most often authors compare 2 portfolios. For exampleDerwall et al.(2005) compares top and bottom 30% of firms ranked by eco-efficiency score. Others create more portfolios to compare not only environmental leaders with environmental laggards but also to study pattern of relationship along ranked companies (de Haan et al. (2012) create 9 portfolios). I consider top 20% as environmental leaders and bottom 20% as environmental laggards. Thus, I create 5 mutually exclusive portfolios based on environ-mental score: one portfolio of environenviron-mental leaders (portfolio 5), one portfolio of environenviron-mental laggards (portfolio 1) and 3 middle portfolios (portfolio 2, 3, and 4).
I create 5 portfolios based on environmental score change to test the second hypothesis. First, I calculate environmental score change as follows:
EScht = (ESt− ESt−1)/ESt, (5)
where ESchtis environmental score change in year t, EStis environmental score in year t, and ESt−1
envi-ronmental score in year previous of year t. Study period of portfolios based on envienvi-ronmental score change is one year shorter (from 2003 to 2012). Then I rank companies by environmental score change and I create 5 portfolios using the same methodology as for creation of portfolios based on environmental score.
After creating portfolios, I calculate monthly portfolio return for each of 5 portfolios as weighted aver-age return of all companies included in a given portfolio, where weight is market capitalization of company. The value weighted average return more accurately capture the total wealth effects experiences by investors than equal-weighted average return (Fama,1998).
5.2. Descriptive statistics of the portfolios
I present descriptive statistics of the portfolios in Table2. Regarding average returns, when portfolios are created based on environmental score environmental leaders portfolio has the lowest average return and portfolio 3 has the highest return. Also portfolio of laggards has higher return than portfolio of leaders. Ziegler et al.(2011) finds similar pattern in average return for portfolio based on Climate Impact Statement presence or absence for European market during 2001 –2006 (average monthly return is 0.0031 of lead-ers portfolio and 0.0089 for laggards portfolio). Opposite descriptive statistics of environmentally screen portfolios are in the study ofDerwall et al.(2005). He finds that, for U.S. companies during 1995 –2003, environmental leaders portfolio has higher monthly average return than the market return, and higher than monthly return of environmental laggards portfolio; 0.0096, 0.009, and 0.0071, respectively.
When portfolios are created based on environmental score change portfolio of laggards has the highest average return and significantly differs from all other portfolios which can be seen in I perform again t-test to determine whether average returns differs across portfolios. Results are in TableE.8.
I perform t-test8to determine if average environmental scores differ significantly across portfolios. In all case p-value for null hypothesis of equal means is 0.000. In addition, in all case p-value for null hypothesis of portfolio x is lower than portfolio x+1 is equal to 1.000 which means that portfolio x has significantly lower environmental score than portfolio x+1. Results are the same when I perform t-test to determine if average environmental score changes differ significantly across portfolios.
5.3. The models
In order to investigate the hypotheses I use two models CAPM and Carhart 4-factor model.Derwall et al. (2005) andZiegler et al.(2011) use also those two models,de Haan et al.(2012) use Carhart 4-factor model but instead of CAPM they use also 5-factor model incorporating risk factor capturing environmental perfor-mance risk (return of “green” portfolios minus return of “non-green” portfolios) in 4-factor model. Firstly, I compare environmental leaders portfolio and environmental laggards portfolio returns using CAPM. This
8I perform t-test for two unpaired samples of independent variables with unequal variances and Welch calculation of degrees of
Table 2: Descriptive Statistics of the portfolios
In this table I show descriptive statistics of the portfolios returns. In PANEL A I show descriptive statistics for portfolios created based on environmental score and in PANEL B descriptive statistic of portfolios created based on environmental score change. Environmental score is environmental score of portfolio calculated as equal-weighted average of environmental score of companies in the portfolio. Environmental score change is calculated as equal-weighted average of environmental score changes of companies in the portfolio. Return is the portfolio return calculated as value-weighted average monthly return of companies in the portfolio. Note: ES stands for environmental score, St. dev stands for standard deviation.
PANEL A ES Return
Mean St. dev Mean St.dev Kurtosis Skewness Minimum Maximum Laggards (portfolio 1) 16.2076 4.9542 0.0027 0.0551 4.3192 -0.6828 -0.1709 0.1455 Portfolio 2 40.3035 12.9186 0.0018 0.0533 5.8901 -1.1749 -0.2223 0.1280 Portfolio 3 68.5784 9.8619 0.0051 0.0483 4.3144 -0.6802 -0.1606 0.1332 Portfolio 4 85.8726 4.7427 0.0030 0.0499 5.0205 -1.1374 -0.1887 0.1085 Leaders (portfolio 5) 93.6864 1.6327 0.0014 0.0512 4.1881 -0.7622 -0.1521 0.1272
PANEL B ES change Return
Mean St. dev Mean St.dev Kurtosis Skewness Minimum Maximum Laggards (portfolio 1) -0.2557 0.1629 0.0035 0.0491 5.1850 -0.9889 -0.1843 0.1127 Portfolio 2 -0.0669 0.0596 0.0002 0.0467 4.1348 -0.8688 -0.1508 0.1071 Portfolio 3 -0.0034 0.0366 0.0015 0.0530 4.3076 -0.5753 -0.1521 0.1479 Portfolio 4 0.0888 0.0787 -0.0011 0.0546 5.1689 -1.1784 -0.2125 0.0997 Leaders (portfolio 5) 0.8341 0.8426 0.0001 0.0530 5.6813 -1.1063 -0.2046 0.1359
one-factor asset pricing model developed bySharpe (1964),Lintner (1965), andMossin (1966) is repre-sented by following equation:
ri,t− rf,t = αi+ βi(rm,t− rf,t)+ εi,t, (6)
where ri,t is the return on the portfolio i in the month t calculated as continuous return from the beginning of the month t to the beginning of the month t+ 1, rf,t is the risk-free rate (EB main refinancing interest rate) in the month t, and rm,t is the market return (STOXX Europe 600 Index) in the month t.
Coefficients αi, and βiare estimated using an ordinary least-squares (OLS) regression. βi can be
inter-preted as a measure of market-risk exposure of portfolio i and αiis an average abnormal return of portfolio
iin excess of market return (Jensen’s alpha). εi,tis an error term with zero identically normally distributed mean and represents the effect of unsystematic and diversifiable risk.
Fama and French (1993) enlarge CAPM by additional two explanatory variables. Their three-factor model incorporates size and value factor which, besides market return, have an explanatory power for excess return of portfolio (ri,t − rf). Four years laterCarhart (1997) creates four-factor model. He incorporates
momentum factor in Fama-French three-factor model. I consider Carhart four-factor model as the second model used in my study.
This model is formulated as the following equation:
ri,t− rf = αi+ β1,i(rm− rf)+ β2,iSMBt+ β3,iHMLt+ β4,iMOMt+ εi,t. (7)
The variables ri,t, rf, and rmremains the same as in CAPM. The coefficient αi and εi,t have the same
interpretations as in CAPM; and parameter β1,i has the same interpretation as parameter βi in CAPM.
HMLt, and MOMt, respectively. All parameters will be estimated by using OLS, as in the case of CAPM.
Variable SMB represents size factor and variable HML represents value factor (book-to-market equity ratio). Both are calculated using same method as is used inFama and French(1993). I create 2 portfolios based on firm size. “Small” firms have market capitalization lower than 50th percentil and “big” firms have market capitalization higher than 50th percentil. Then I create 2 portfolios based on book-to-market equity ratio. “Value” firms have book-to-market equity ratio higher than 70th percentil and “growth” firms have book-to-market equity ratio lower than 30th percentil. Then I create portfolio of “small-value” firms, “small-growth” firms, “big-value” firms, and “big-growth” firms and I calculated portfolio returns as value-weighted average return. Then I calculate monthly returns of portfolio including “small” firms as equal-weighted average return of portfolio “small-value” and “small-growth”. In the same way I calculate monthly returns of portfolio including “big” firms, then including “value” and portfolio including “growth” firms. Finally I calculate SMBt factor as the difference between the return of portfolio of “small” firms and the
return of portfolio of “big” firms and HMLtfactor as the difference between the return of portfolio of “value”
firms and the return of portfolio of “growth” firms.
The last variable, MOMt, capture the one-year momentum effect. I follow methodology of Carhart
(1997). First, I create portfolio of last year “winners” including firms with stock return higher than 70th percentil during last year (Carhart (1997) consider last year as period from month t-12 to month t-2 – 11 months) and I also create portfolio of last year “losers” (portfolio including firms with last year return lower than 30th percentil). I calculate the return of portfolio of “winners” and the return of portfolio of “losers” as value-weighted average return of all including companies. Descriptive statistics and correlations of size portfolio, value portfolio, and momentum portfolio are in TableF.9. Are not high enough to lead to multicollinearity problem (the highest correlation in magnitude is -0.6066 between market premium and momentum factor).
5.4. Sensitivity analysis
To test the robustness of results obtained from regressions of CAPM and 4-factor model I perform 3 sensitivity analyzes regarding the structural break in dataset, the region and industry effect of environmental score, and number of portfolios I create.
Recently we encountered a huge financial crisis which shook financial markets and influenced financial figures and stock prices of most of the listed companies. Therefore, I include in my regressions dummy vari-able for financial crisis to test whether a structural break is present in my dataset. Generally, the bankruptcy of Lehman Brothers on September 15, 2008 is considered to be the beginning of financial crisis. Thus the dummy variable for crisis takes value of 1 for September 2008 and all following months, and value of 0 for August 2008 and all previous months.
(5 firms), Hungary (5 firms), and Poland (19 firms) I exclude those countries from sample for this sensi-tivity analysis. Then my sample includes 10 industries and 17 countries and consequently the number of dependent industry-country groups is to high (170). Thus, I create country and industry clusters – regions and sectors. I perform a cluster analysis based on the correlations which capture evolution of environmen-tal score and on average portfolio environmenenvironmen-tal score which capture average level of environmenenvironmen-tal score during study period.
Then, I adjust firm environmental score for region-sector average using following method. At the begin-ning I calculate equal-weighted average environmental score for each region-sector in February each year. I calculate adjusted environmental score of all firms as firm environmental score minus average environ-mental score of region-sector this firm belongs. Then I rank companies by adjusted environenviron-mental score, I calculate each year quintiles, and I create 5 mutually exclusive portfolios (following the same methodol-ogy as for non-adjusted environmental score). Similarly I calculate environmental score change for each region-sector and I adjust firm‘s environmental score change. Then I create 5 portfolios based on adjusted environmental score change. Finally, I perform regression of CAPM and 4-factor model. The correlation of average environmental score of countries and of industries, as well as the final redistribution of industries into sectors and countries into regions are in TableG.10, TableG.11, and TableG.12, respectively.
Third, I test whether number of created portfolios has an influence on results. As I primarily create 5 mutually exclusive portfolios I test for 4 and 6 portfolios created. In case of 4 portfolios I consider top 25% companies as environmental leaders and bottom 25% as environmental laggards and I create 2 middle portfolios. From this point methodology of portfolios creation is the same as in case of 5 portfolios created. In case of 6 portfolios I consider to 16% companies as environmental leaders and bottom 16% as environmental laggards and I create 4 middle portfolios (each with range of 17 percentils). Again, from this point I follow methodology used in case of 5 portfolios.
5.5. Heteroscedasticity test
I perform Breusch-Pagan / Cook-Weisberg test for heteroscedasticity for both models, CAPM and 4-factor model, and for portfolios created based on environmental score level and environmental score changes. For each case the probability of homoscedasticity is p= 0 which means that standard errors are not homoscedastic. Although heteroscedasticity does not cause biased coefficient estimates it can lead to biased estimates of the variance of the coefficients and consequently biased standard errors (Brooks,2008). Therefore I use robust standard errors for all regressions which address the problem of heteroscedasticity (White,1980).
5.6. Downside risk measure and risk-return trade-off measures
Similarly to Blank and Daniel(2002) and Filbeck and Gorman(2004) I study Sharpe ratio which is a risk-return trade-off measure. Sharpe ratio represents excess return over a unit of risk using standard deviation as measure of risk.This measure allow comparison of across portfolios not only based on return but also covering risk factor. Sharpe ratio (Sharpe,1966) is calculated as following:
Sharpe ratio= E(rp,t− rz,t)
σ , (8)
is usually risk free return, market return or average return of given portfolio. I use two first options. Thus in my study rz,t stands for or risk free rate either market return in the month t.
In addition to Sharpe ratio I investigate also Sortino ratio which differentiates from Sharpe ratio by using downside risk measure and allow capturing only investor perspective relevant risk of losses. To measure downside risk I use semi standard deviation (SSD) (Sortino and van der Meer, 1991). The essential of downside risk measures is to catch risk below a threshold value:
SSD= v t 1 n × n X i=1 min(0, rz,t− rp,t)2, (9)
where rz,t and rp,t are defined as in equation 8. rp,t is portfolio p return in month t, n is number of observations. After calculation of semi-deviation I calculate Sortino ratio (Nawrocki,1999) as:
Sortino ratio= E(rp,t− rz,t)
SSD , (10)
where all variable defined as in equation (9) and equation (8).
6. Results
6.1. CAPM results
In Table 3, I present results of regressed monthly excess return of all 5 portfolios created based on environmental score. First of all, this results support CAPM model as β, representing market factor, is significant at 1% confidence interval for 5 portfolios with approximately same value varying from 0.962 to 1.077. Jensen‘s alpha has very interesting pattern across the portfolios. First, portfolio 1 – environmental laggards – and portfolio 2 have negative α of - 0.00178 and -0.00262, respectively, but both coefficients are statistically insignificant. Then the coefficient for portfolio 3 is positive but very small (0.00075) and also statistically insignificant. Portfolio 4 has negative Jensen‘s alpha which is significant with confidence interval of 10%. Contrary to my expectations, Jensen‘s alpha of portfolio 5 is negative with value -0.00339 and significant with confidence interval of 1%.
When comparing Jensen‘s alpha of the environmental leaders portfolio and the environmental laggards portfolio we see that Jensen‘s alpha is lower for the environmental leaders portfolio and, in addition, it is also significant while Jensen‘s alpha for the environmental laggards portfolio is not significantly different from zero. This indicates that the portfolio of environmental laggards does not differ from the market but the portfolio of environmental leaders underperforms the market. However, we cannot reject that Jensen‘s alpha of portfolio 1 differs from Jensen‘s alpha of portfolio 5 as the null hypothesis of the test of equal constants cannot be rejected (p-value= 0.421). Thus, I cannot reject HA0 which means that the return of
portfolio of environmental leaders does not significantly differ from the return of portfolio of environmental laggards.
I reject hypothesis HB0as Jensen‘s alpha of portfolio 5 is statistically significant. However, I expected
Table 3: Results of CAPM when portfolios are created based on environmental score
In this table I show coefficients of CAPM regression. Portfolios are created based on environmental score. I regress monthly excess return of portfolio in month t using equation αi+βi(rm,t−rf,t)+εi,t, where α is Jensen‘s alpha, β is market factor, (rm,t−rf,t) is market
premium (calculated as logarithmic market return minus risk free rate in month t), and εi,tis error term. Test of equal constant tests
the null hypothesis H0: Jensen‘s alpha of portfolio 1= Jensen‘s alpha of portfolio 5.
Laggards (portfolio 1) Portfolio 2 Portfolio 3 Portfolio 4 Leaders (portfolio 5) Market factor 1.077*** 1.040*** 0.962*** 0.996*** 1.023*** (0.0405) (0.0576) (0.0385) (0.0342) (0.0243) Jensen‘s alpha -0.00178 -0.00262 0.00075 -0.00179* -0.00339*** (0.0017) (0.0019) (0.0012) (0.0010) (0.0010) Observations 122 122 122 122 122 R-squared 0.882 0.866 0.928 0.954 0.958
Test of equal constants (p-value) 0.408
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Ziegler et al.(2011) also find that the portfolio of environmental leaders (in terms of present climate im-pact statement) has significant negative Jensen‘s alpha (-0.71%) and the portfolio of environmental laggards (in terms of absent climate impact statement) has slightly negative Jensen‘s alpha (-0.06%) but statistically insignificant. However,Derwall et al.(2005) for U.S. market andZiegler et al.(2007) for European market find results opposite to mine. The findings ofDerwall et al.(2005) indicate that the portfolio of environ-mental leaders outperforms the portfolio of environenviron-mental laggards. In his results Jensen‘s alpha in CAPM regression is significantly positive for the former (1.29% annualized) and negative (-1.76% annualized) but insignificant for the latter. Ziegler et al. (2007) use multiple regression method and his results suggest rather positive relationship between environmental and financial performance for European market (212 companies from 13 countries).
Results of CAPM model for the portfolios based on environmental score change are in Table4. First, CAPM model is again confirmed as β1is significant with confidence interval of 1% for all 5 portfolios and
has approximately same value across portfolios (from 0.919 to 1.04) and also comparing to coefficient of market factor in Table3. Comparing to results from Table3, Jensen‘s alpha across portfolios has similar pattern. Portfolio 1 and 2 have slightly positive Jensen‘s alpha but statistically not different from zero. Portfolios 3, 4, and 5 have negative Jensen‘s alpha and all statistically significant (with confidence interval of 5% for portfolio 3, and with confidence interval of 1% for portfolio 4 and 5).
Based on the test of equal constants (p-value= 0.185) I cannot reject that Jensen‘s alpha of portfolio 1 and portfolio 5 are equal and thus hypothesis HC0 cannot be rejected. This means that the portfolio
of firms with increasing environmental performance does not have significantly different return from the portfolio of firms with decreasing environmental score. But it is worth to mention that p-value is closer to 0 than to 1 so it could indicate some difference between those coefficients suggesting that the portfolio of firms with increasing environmental performance slightly underperforms the portfolio of firms with strongly decreasing environmental score.
I cannot reject hypothesis HD0with confidence interval of 10% which means that the return of portfolio
Table 4: Results of CAPM when portfolios are created based on environmental score change
In this table I show coefficients of CAPM regression. Portfolios are created based on environmental score change. I regress monthly excess return of portfolio in month t using equation αi+ βi(rm,t− rf,t)+ εi,t, where α is Jensen‘s alpha, β is market factor, (rm,t− rf,t)
is market premium (calculated as logarithmic market return minus risk free rate in month t) and εi,tis error term. Test of equal
constant tests the null hypothesis H0: Jensen‘s alpha of portfolio 1= Jensen‘s alpha of portfolio 5.
Laggards (portfolio 1) Portfolio 2 Portfolio 3 Portfolio 4 Leaders (portfolio 5) Market factor 0.942*** 0.942*** 0.919*** 1.040*** 1.028*** (0.0520) (0.0520) (0.0353) (0.0274) (0.0510) Jensen‘s alpha 0.00020 0.00020 -0.00254** -0.00199* -0.00270* (0.0016) (0.0016) (0.0011) (0.0010) (0.0016) Observations 110 110 110 110 110 R-squared 0.892 0.892 0.940 0.956 0.911
Test of equal constants (p-value) 0.185
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
with increasing environmental score has lower return that the market.
In CAPM framework, the results indicates that it does not pay to invest in “green” companies on Eu-ropean market because the portfolio including top 20% firms in environmental performance significantly underperforms the market. I addition, it does not pay to invest in the portfolio of companies improving their environmental performance as this underperforms the market, too. On the other hand, it seems that the investment in the portfolio of environmental laggards, and in the portfolio of firms with strongly decreasing environmental performance leads to the return comparable to the market return. The results indicate that the relationship between environmental performance and financial performance is negative. This pattern is observable in both cases, when the portfolios are created based on environmental score and when the portfolios are created based on environmental score change. Thus, the “traditionalist” view seems to be hold and it appears that market links higher environmental performance to higher costs which lower the shareholders‘ value (Walley and Whitehead, 1994). Alternative explanation may rely in a non-financial compensation (“good feeling” from environment protection) for investors who like to invest in “green” portfolios, or in higher financial reward for investing in environment unfriendly portfolios (to offset “bad feeling” from increased environmental load from such a investment).
6.2. Carhart 4-factor model results
Table 5: Results of Carhart 4-factor model when portfolios are created based on environmental score
In this table I show coefficients of Carhart 4-factor model regression. Portfolios are created based on environmental score. I regress monthly excess return of portfolio in month t using equation ri,t− rf = αi+ β1,i(rm− rf)+ β2,iSMBt+ β3,iHMLt+ β4,iMOMt+ εi,t,
where α is Jensen‘s alpha, β1 is market factor, β2 is size factor,β3 is value factor,β4 is momentum factor, (rm,t− rf,t) is market
premium (calculated as logarithmic market return minus risk free rate in month t), SMBt, HMLt, MOMt is the logarithmic return
on the mimicking portfolio representing size risk, value risk, and momentum risk, respectively, and εi,tis error term. Test of equal
constant tests the null hypothesis H0: Jensen‘s alpha of portfolio 1= Jensen‘s alpha of portfolio 5.
Laggards (portfolio 1) Portfolio 2 Portfolio 3 Portfolio 4 Leaders (portfolio 5) Market factor 0.994*** 1.019*** 0.961*** 1.008*** 1.001*** (0.0416) (0.0439) (0.0349) (0.0371) (0.0292) Size factor 0.467*** 0.567*** 0.331*** 0.047 -0.095* (0.0808) (0.0824) (0.0474) (0.0381) (0.0538) Value factor -0.002 -0.207** -0.157*** 0.077 0.076 (0.0729) (0.0867) (0.0542) (0.0483) (0.0507) Momentum factor 0.021 -0.008 -0.017 0.104*** -0.009 (0.0440) (0.0520) (0.0288) (0.0330) (0.0315) Jensen‘s alpha -0.00295* -0.00355** 0.00038 -0.00328*** -0.00310*** (0.0017) (0.0017) (0.0011) (0.0010) (0.0012) Observations 122 122 122 122 122 R-squared 0.917 0.917 0.950 0.959 0.961
Test of equal constants (p-value) 0.938
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Regarding value factor, only coefficients for portfolios 2 and 3 are negative and significant. Coefficients for portfolios 1, 4 and 5 are not statistically different from zero. Similarly, momentum factor coefficient is significant and positive only in regression for portfolio 4 and for all other portfolios this coefficient is not significant.
Looking on Jensen‘s alphas, an interesting pattern can be seen. First, for portfolios 1, 2, 4, and 5 Jensen‘s alpha is negative and with approximately same magnitude (from -0.00355 to -0.00295). For portfolio 1 and 2 Jensen‘s alpha is significant with confidence interval of 10% and 5%, respectively; and for portfolios 4, and 5 with confidence interval of 1%. Second, in case of portfolio 3, Jensen‘s alpha is slightly positive but insignificant. Thus it seems that only the portfolio of companies with average environmental performance leads to at least market return.
The result of t-test of equal constants in the last row of Table5indicates that, with confidence interval of 10%, Jensen‘s alphas of portfolio 1 and 5 are equal. This mean that I cannot reject HA0and that the return of
the environmental leaders portfolio does not differ from the return of the environmental laggards portfolio. On the other hand, I reject HB0 as Jensen‘s alpha for the environmental leaders portfolio is statistically
alpha even though it is higher for the environmental leaders portfolio. The latter find higher and more significant Jensen‘s alpha for the environmental laggards portfolios.
In Table6I show the results of 4-factor model for the portfolios based on environmental score changes. As for previous 3 regressions, market factor coefficient is significant for all 5 portfolios at the confidence interval of 1% and takes approximately same values across the portfolios (from 0.940 to 1.070).
Table 6: Results of Carhart 4-factor model when portfolios are created based on environmental score change
In this table I show coefficients of Carhart 4-factor model regression. Portfolios are created based on environmental score change. I regress monthly excess return of portfolio in month t using equation ri,t−rf = αi+β1,i(rm−rf)+β2,iSMBt+β3,iHMLt+β4,iMOMt+εi,t,
where α is Jensen‘s alpha, β1 is market factor, β2 is size factor,β3 is value factor,β4 is momentum factor, (rm,t− rf,t) is market
premium (calculated as logarithmic market return minus risk free rate in month t), SMBt, HMLt, MOMt is the logarithmic return
on the mimicking portfolio representing size risk, value risk, and momentum risk, respectively, and εi,tis error term. Test of equal constant tests the null hypothesis H0: Jensen‘s alpha of portfolio 1= Jensen‘s alpha of portfolio 5.
Laggards (portfolio 1) Portfolio 2 Portfolio 3 Portfolio 4 Leaders (portfolio 5) Market factor 0.962*** 0.940*** 0.972*** 1.070*** 1.021*** (0.0367) (0.0363) (0.0327) (0.0459) (0.0438) Size factor 0.427*** 0.016 -0.039 0.065 0.375*** (0.0748) (0.0656) (0.0591) (0.0987) (0.0821) Value factor -0.106 0.097 0.113** -0.069 -0.195*** (0.0759) (0.0599) (0.0491) (0.0852) (0.0740) Momentum factor 0.099** 0.136*** -0.036 -0.020 -0.055 (0.0447) (0.0347) (0.0285) (0.0521) (0.0368) Jensen‘s alpha -0.00147 -0.00439*** -0.00141 -0.00428** -0.00226 (0.0013) (0.0011) (0.0010) (0.0016) (0.0014) Observations 110 110 110 110 110 R-squared 0.932 0.949 0.963 0.932 0.936
Test of equal constants (p-value) 0.680
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Size factor coefficient is significantly positive in case of portfolio 1 and 5. Remaining 3 portfolios are not significantly exposed to this factor with confidence interval of 1%. Exposure to value factor differs across the portfolios. While portfolios 1, 2 and 4 have insignificant value factor coefficients, portfolios 3 and 5 have significant coefficients. But coefficient for portfolio 3 is positive with value of 0.113 and coefficient for portfolio 5 is negative with value of -0.195. This indicates that the portfolio of companies with increasing environmental performance is biased towards growth firms while portfolio 3 including firms with small decrease in environmental score (based average value of environmental score change in Table2) is biased towards value firms. Momentum factor has significant influence on portfolio 2 where coefficient takes value of 0.136 and is significant with confidence interval of 1%. For other portfolios this risk factor is statistically insignificant.
